[Japanese | Thesis | Researches in Minoh Lab | Minoh Lab]

Shape Modeling of String-like Flexible Object from Real Observation for its Manipulation in Virtual World

This paper discusses modeling the shape of a string-like flexible object, based on observation of a real string in various shape, for visualizing it in a virtual world.

Recently, the technology of Virtual Reality (VR) is becoming popular and is improved day by day. It becomes significantly important to realizing the visual and haptic feedback in virtual world. For this purpose, we need virtualization --- creating virtual objects that behave like real objects. In conventional VR applications, the objects manipulated in virtual world were mostly rigid objects. Those objects can be virtualized by realizing comparatively few properties. On the other hand, virtualization of flexible object is required for new applications with manipulating various object in a virtual world. It is difficult to virtualize flexible object because they have more complicated geometric and mechanical properties in various conditions than rigid ones. We should virtualize flexible objects so that virtualized objects can realize those properties.

In the previous work on virtualization of flexible object, cloth simulation for computer graphics (CG) has been discussed. In that work, the parameters for the object model have to be modified by the designer so that the behavior realized by the model is acceptable as a cloth. Under this approach, it is troublesome work to find appropriate values for the parameters especially to realize the differences in the behavior of various clothes. In recent work, it has been proposed to model clothes based on their physical properties measured by a special device based on KES (Kawabata's Evaluation System). This method, however, requires us to use such a special device and does not guarantee the model reflects its visual and haptic feedback given to users.

We take a task-oriented standpoint for virtualizing a real object, based on the fact that each virtual object is employed only for generating visual and haptic feedback when we manipulate it. From this standpoint, we aim to virtualize an object so that it gives the same feedback as the real object based on the behavior of the object observed by sensors including a camera. The virtual object modeled under this approach only guarantees to behave like the real object in the same conditions in which the real object has been observed. If we observe the behavior of the real object under all possible conditions during its manipulation, we can obtain the model of the object that can realize the behavior for its manipulation in virtual world.

This paper focuses on the visual feedback of a flexible object like a string. The objective is to obtain a model of a string to visualize static shapes of a real string under this approach. We propose the method for finding appropriate the parameters' values of the model of a string from observation of its shape by a camera.

Our method employs mass-spring model, which is often used in cloth simulation, for the model of string. It is constituted by springs that represent the properties of a real string against stretching and bending. The parameters of the model including the natural length of each string, are modified so that the shape of the model fits to the static shape of a real string observed by a camera. The values of the parameters are updated based on energy minimization by steepest descent method. The energy represents the difference of the model's shape from the observed shape together with the internal energy of the model. With the parameters obtained as the result of this energy minimization, the model can visualize each observed shape of the real string.

Modeling a string by our method is evaluated by the three kinds of experiments. First, a rubber string is modeled from a single observed shape, and it is confirmed that the resultant model can properly take the observed shape. Second, our method is applied to model the two different kinds of strings : a rubber string and a chain. The parameters obtained by this method for these different kinds of strings reflects the difference of the properties of these strings. Finally, a rubber string is modeled from a various observed shapes. The model can realize unobserved shapes, as well as observed ones, of the real string. These results show the validity of our method.

Go back to Thesis Page